PlanOpticon

Add Ollama provider for fully offline video analysis Adds Ollama as a fourth LLM provider, enabling fully offline operation with no API keys required. Uses Ollama's OpenAI-compatible API (zero new dependencies). Auto-detects running Ollama server and falls back to it when no cloud API keys are configured. - New OllamaProvider class with chat, vision, and model discovery - ProviderManager auto-fallback to Ollama as last resort - Vision model detection for llava, moondream, etc. - CLI --provider ollama option on all commands - Updated docs, architecture guide, and CLI reference - 7 new tests for Ollama provider integration

lmata 2026-02-16 22:44 trunk

Commit 139b64b39f1bd0daec84e957d55b085a9c1c6eea19c65cf11bcf58f5dc7437cd

Parent 8ade7827de290d0…

12 files changed +3 -3 +2 -1 +38 -5 +3 -3 +6 -3 +20 -1 +3 +91 -3 +9 -7 +12 +27 -2 +109

~ README.md ~ docs/architecture/overview.md ~ docs/architecture/providers.md ~ docs/cli-reference.md ~ docs/getting-started/configuration.md ~ docs/getting-started/installation.md ~ docs/getting-started/quickstart.md ~ tests/test_providers.py ~ video_processor/cli/commands.py ~ video_processor/providers/discovery.py ~ video_processor/providers/manager.py + video_processor/providers/ollama_provider.py

M README.md

+3 -3

		--- README.md
		+++ README.md
		@@ -6,15 +6,15 @@
6	6	[![License](https://img.shields.io/github/license/ConflictHQ/PlanOpticon)](LICENSE)
7	7	[![Docs](https://img.shields.io/badge/docs-planopticon.dev-blue)](https://planopticon.dev)
8	8
9	9	AI-powered video analysis and knowledge extraction.
10	10
11		-PlanOpticon processes video recordings into structured knowledge — transcripts, diagrams, action items, key points, and knowledge graphs. It auto-discovers available models across OpenAI, Anthropic, and Gemini, and produces rich multi-format output.
	11	+PlanOpticon processes video recordings into structured knowledge — transcripts, diagrams, action items, key points, and knowledge graphs. It auto-discovers available models across OpenAI, Anthropic, Gemini, and Ollama, and produces rich multi-format output.
12	12
13	13	## Features
14	14
15		-- Multi-provider AI — Auto-discovers and routes to the best available model across OpenAI, Anthropic, and Google Gemini
	15	+- Multi-provider AI — Auto-discovers and routes to the best available model across OpenAI, Anthropic, Google Gemini, and Ollama (fully offline)
16	16	- Smart frame extraction — Change detection for transitions + periodic capture for slow-evolving content (document scrolling, screen shares)
17	17	- People frame filtering — OpenCV face detection automatically removes webcam/video conference frames, keeping only shared content
18	18	- Diagram extraction — Vision model classification detects flowcharts, architecture diagrams, charts, and whiteboards
19	19	- Knowledge graphs — Extracts entities and relationships, builds and merges knowledge graphs across videos
20	20	- Action item detection — Finds commitments, tasks, and follow-ups with assignees and deadlines
		@@ -65,11 +65,11 @@
65	65
66	66	### Requirements
67	67
68	68	- Python 3.10+
69	69	- FFmpeg (`brew install ffmpeg` / `apt install ffmpeg`)
70		-- At least one API key: `OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, or `GEMINI_API_KEY`
	70	+- At least one API key (`OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, or `GEMINI_API_KEY`) or [Ollama](https://ollama.com) running locally
71	71
72	72	## Output Structure
73	73
74	74	```
75	75	output/
76	76

	--- README.md
	+++ README.md
	@@ -6,15 +6,15 @@
6	[![License](https://img.shields.io/github/license/ConflictHQ/PlanOpticon)](LICENSE)
7	[![Docs](https://img.shields.io/badge/docs-planopticon.dev-blue)](https://planopticon.dev)
8
9	AI-powered video analysis and knowledge extraction.
10
11	PlanOpticon processes video recordings into structured knowledge — transcripts, diagrams, action items, key points, and knowledge graphs. It auto-discovers available models across OpenAI, Anthropic, and Gemini, and produces rich multi-format output.
12
13	## Features
14
15	- Multi-provider AI — Auto-discovers and routes to the best available model across OpenAI, Anthropic, and Google Gemini
16	- Smart frame extraction — Change detection for transitions + periodic capture for slow-evolving content (document scrolling, screen shares)
17	- People frame filtering — OpenCV face detection automatically removes webcam/video conference frames, keeping only shared content
18	- Diagram extraction — Vision model classification detects flowcharts, architecture diagrams, charts, and whiteboards
19	- Knowledge graphs — Extracts entities and relationships, builds and merges knowledge graphs across videos
20	- Action item detection — Finds commitments, tasks, and follow-ups with assignees and deadlines
	@@ -65,11 +65,11 @@
65
66	### Requirements
67
68	- Python 3.10+
69	- FFmpeg (`brew install ffmpeg` / `apt install ffmpeg`)
70	- At least one API key: `OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, or `GEMINI_API_KEY`
71
72	## Output Structure
73
74	```
75	output/
76

	--- README.md
	+++ README.md
	@@ -6,15 +6,15 @@
6	[![License](https://img.shields.io/github/license/ConflictHQ/PlanOpticon)](LICENSE)
7	[![Docs](https://img.shields.io/badge/docs-planopticon.dev-blue)](https://planopticon.dev)
8
9	AI-powered video analysis and knowledge extraction.
10
11	PlanOpticon processes video recordings into structured knowledge — transcripts, diagrams, action items, key points, and knowledge graphs. It auto-discovers available models across OpenAI, Anthropic, Gemini, and Ollama, and produces rich multi-format output.
12
13	## Features
14
15	- Multi-provider AI — Auto-discovers and routes to the best available model across OpenAI, Anthropic, Google Gemini, and Ollama (fully offline)
16	- Smart frame extraction — Change detection for transitions + periodic capture for slow-evolving content (document scrolling, screen shares)
17	- People frame filtering — OpenCV face detection automatically removes webcam/video conference frames, keeping only shared content
18	- Diagram extraction — Vision model classification detects flowcharts, architecture diagrams, charts, and whiteboards
19	- Knowledge graphs — Extracts entities and relationships, builds and merges knowledge graphs across videos
20	- Action item detection — Finds commitments, tasks, and follow-ups with assignees and deadlines
	@@ -65,11 +65,11 @@
65
66	### Requirements
67
68	- Python 3.10+
69	- FFmpeg (`brew install ffmpeg` / `apt install ffmpeg`)
70	- At least one API key (`OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, or `GEMINI_API_KEY`) or [Ollama](https://ollama.com) running locally
71
72	## Output Structure
73
74	```
75	output/
76

M docs/architecture/overview.md

+2 -1

		--- docs/architecture/overview.md
		+++ docs/architecture/overview.md
		@@ -43,10 +43,11 @@
43	43	├── providers/ # AI provider abstraction
44	44	│ ├── base.py # BaseProvider ABC
45	45	│ ├── openai_provider.py
46	46	│ ├── anthropic_provider.py
47	47	│ ├── gemini_provider.py
	48	+│ ├── ollama_provider.py # Local Ollama (offline)
48	49	│ ├── discovery.py # Auto-model-discovery
49	50	│ └── manager.py # ProviderManager routing
50	51	├── utils/
51	52	│ ├── json_parsing.py # Robust LLM JSON parsing
52	53	│ ├── rendering.py # Mermaid + chart rendering
		@@ -60,8 +61,8 @@
60	61
61	62	## Key design decisions
62	63
63	64	- Pydantic everywhere — All structured data uses pydantic models for validation and serialization
64	65	- Manifest-driven — Every run produces `manifest.json` as the single source of truth
65		-- Provider abstraction — Single `ProviderManager` wraps OpenAI, Anthropic, Gemini behind a common interface
	66	+- Provider abstraction — Single `ProviderManager` wraps OpenAI, Anthropic, Gemini, and Ollama behind a common interface
66	67	- No hardcoded models — Model lists come from API discovery
67	68	- Screengrab fallback — When extraction fails, save the frame as a captioned screenshot
68	69

	--- docs/architecture/overview.md
	+++ docs/architecture/overview.md
	@@ -43,10 +43,11 @@
43	├── providers/ # AI provider abstraction
44	│ ├── base.py # BaseProvider ABC
45	│ ├── openai_provider.py
46	│ ├── anthropic_provider.py
47	│ ├── gemini_provider.py

48	│ ├── discovery.py # Auto-model-discovery
49	│ └── manager.py # ProviderManager routing
50	├── utils/
51	│ ├── json_parsing.py # Robust LLM JSON parsing
52	│ ├── rendering.py # Mermaid + chart rendering
	@@ -60,8 +61,8 @@
60
61	## Key design decisions
62
63	- Pydantic everywhere — All structured data uses pydantic models for validation and serialization
64	- Manifest-driven — Every run produces `manifest.json` as the single source of truth
65	- Provider abstraction — Single `ProviderManager` wraps OpenAI, Anthropic, Gemini behind a common interface
66	- No hardcoded models — Model lists come from API discovery
67	- Screengrab fallback — When extraction fails, save the frame as a captioned screenshot
68

	--- docs/architecture/overview.md
	+++ docs/architecture/overview.md
	@@ -43,10 +43,11 @@
43	├── providers/ # AI provider abstraction
44	│ ├── base.py # BaseProvider ABC
45	│ ├── openai_provider.py
46	│ ├── anthropic_provider.py
47	│ ├── gemini_provider.py
48	│ ├── ollama_provider.py # Local Ollama (offline)
49	│ ├── discovery.py # Auto-model-discovery
50	│ └── manager.py # ProviderManager routing
51	├── utils/
52	│ ├── json_parsing.py # Robust LLM JSON parsing
53	│ ├── rendering.py # Mermaid + chart rendering
	@@ -60,8 +61,8 @@
61
62	## Key design decisions
63
64	- Pydantic everywhere — All structured data uses pydantic models for validation and serialization
65	- Manifest-driven — Every run produces `manifest.json` as the single source of truth
66	- Provider abstraction — Single `ProviderManager` wraps OpenAI, Anthropic, Gemini, and Ollama behind a common interface
67	- No hardcoded models — Model lists come from API discovery
68	- Screengrab fallback — When extraction fails, save the frame as a captioned screenshot
69

M docs/architecture/providers.md

+38 -5

		--- docs/architecture/providers.md
		+++ docs/architecture/providers.md
		@@ -9,40 +9,73 @@
9	9	\| Provider \| Chat \| Vision \| Transcription \|
10	10	\|----------\|------\|--------\|--------------\|
11	11	\| OpenAI \| GPT-4o, GPT-4 \| GPT-4o \| Whisper-1 \|
12	12	\| Anthropic \| Claude Sonnet/Opus \| Claude Sonnet/Opus \| — \|
13	13	\| Google Gemini \| Gemini Flash/Pro \| Gemini Flash/Pro \| Gemini Flash \|
	14	+\| Ollama (local) \| Any installed model \| llava, moondream, etc. \| — (use local Whisper) \|
	15	+
	16	+## Ollama (offline mode)
	17	+
	18	+[Ollama](https://ollama.com) enables fully offline operation with no API keys required. PlanOpticon connects via Ollama's OpenAI-compatible API.
	19	+
	20	+```bash
	21	+# Install and start Ollama
	22	+ollama serve
	23	+
	24	+# Pull a chat model
	25	+ollama pull llama3.2
	26	+
	27	+# Pull a vision model (for diagram analysis)
	28	+ollama pull llava
	29	+```
	30	+
	31	+PlanOpticon auto-detects Ollama when it's running. To force Ollama:
	32	+
	33	+```bash
	34	+planopticon analyze -i video.mp4 -o ./out --provider ollama
	35	+```
	36	+
	37	+Configure a non-default host via `OLLAMA_HOST`:
	38	+
	39	+```bash
	40	+export OLLAMA_HOST=http://192.168.1.100:11434
	41	+```
14	42
15	43	## Auto-discovery
16	44
17		-On startup, `ProviderManager` checks which API keys are configured and queries each provider's API to discover available models:
	45	+On startup, `ProviderManager` checks which API keys are configured, queries each provider's API, and checks for a running Ollama server to discover available models:
18	46
19	47	```python
20	48	from video_processor.providers.manager import ProviderManager
21	49
22	50	pm = ProviderManager()
23		-# Automatically discovers models from all configured providers
	51	+# Automatically discovers models from all configured providers + Ollama
24	52	```
25	53
26	54	## Routing preferences
27	55
28	56	Each task type has a default preference order:
29	57
30	58	\| Task \| Preference \|
31	59	\|------\|-----------\|
32		-\| Vision \| Gemini Flash → GPT-4o → Claude Sonnet \|
33		-\| Chat \| Claude Sonnet → GPT-4o → Gemini Flash \|
34		-\| Transcription \| Whisper-1 → Gemini Flash \|
	60	+\| Vision \| Gemini Flash → GPT-4o → Claude Sonnet → Ollama \|
	61	+\| Chat \| Claude Sonnet → GPT-4o → Gemini Flash → Ollama \|
	62	+\| Transcription \| Local Whisper → Whisper-1 → Gemini Flash \|
	63	+
	64	+Ollama acts as the last-resort fallback — if no cloud API keys are set but Ollama is running, it is used automatically.
35	65
36	66	## Manual override
37	67
38	68	```python
39	69	pm = ProviderManager(
40	70	vision_model="gpt-4o",
41	71	chat_model="claude-sonnet-4-5-20250929",
42	72	provider="openai", # Force a specific provider
43	73	)
	74	+
	75	+# Or use Ollama for fully offline processing
	76	+pm = ProviderManager(provider="ollama")
44	77	```
45	78
46	79	## BaseProvider interface
47	80
48	81	All providers implement:
49	82

	--- docs/architecture/providers.md
	+++ docs/architecture/providers.md
	@@ -9,40 +9,73 @@
9	\| Provider \| Chat \| Vision \| Transcription \|
10	\|----------\|------\|--------\|--------------\|
11	\| OpenAI \| GPT-4o, GPT-4 \| GPT-4o \| Whisper-1 \|
12	\| Anthropic \| Claude Sonnet/Opus \| Claude Sonnet/Opus \| — \|
13	\| Google Gemini \| Gemini Flash/Pro \| Gemini Flash/Pro \| Gemini Flash \|




























14
15	## Auto-discovery
16
17	On startup, `ProviderManager` checks which API keys are configured and queries each provider's API to discover available models:
18
19	```python
20	from video_processor.providers.manager import ProviderManager
21
22	pm = ProviderManager()
23	# Automatically discovers models from all configured providers
24	```
25
26	## Routing preferences
27
28	Each task type has a default preference order:
29
30	\| Task \| Preference \|
31	\|------\|-----------\|
32	\| Vision \| Gemini Flash → GPT-4o → Claude Sonnet \|
33	\| Chat \| Claude Sonnet → GPT-4o → Gemini Flash \|
34	\| Transcription \| Whisper-1 → Gemini Flash \|


35
36	## Manual override
37
38	```python
39	pm = ProviderManager(
40	vision_model="gpt-4o",
41	chat_model="claude-sonnet-4-5-20250929",
42	provider="openai", # Force a specific provider
43	)



44	```
45
46	## BaseProvider interface
47
48	All providers implement:
49

	--- docs/architecture/providers.md
	+++ docs/architecture/providers.md
	@@ -9,40 +9,73 @@
9	\| Provider \| Chat \| Vision \| Transcription \|
10	\|----------\|------\|--------\|--------------\|
11	\| OpenAI \| GPT-4o, GPT-4 \| GPT-4o \| Whisper-1 \|
12	\| Anthropic \| Claude Sonnet/Opus \| Claude Sonnet/Opus \| — \|
13	\| Google Gemini \| Gemini Flash/Pro \| Gemini Flash/Pro \| Gemini Flash \|
14	\| Ollama (local) \| Any installed model \| llava, moondream, etc. \| — (use local Whisper) \|
15
16	## Ollama (offline mode)
17
18	[Ollama](https://ollama.com) enables fully offline operation with no API keys required. PlanOpticon connects via Ollama's OpenAI-compatible API.
19
20	```bash
21	# Install and start Ollama
22	ollama serve
23
24	# Pull a chat model
25	ollama pull llama3.2
26
27	# Pull a vision model (for diagram analysis)
28	ollama pull llava
29	```
30
31	PlanOpticon auto-detects Ollama when it's running. To force Ollama:
32
33	```bash
34	planopticon analyze -i video.mp4 -o ./out --provider ollama
35	```
36
37	Configure a non-default host via `OLLAMA_HOST`:
38
39	```bash
40	export OLLAMA_HOST=http://192.168.1.100:11434
41	```
42
43	## Auto-discovery
44
45	On startup, `ProviderManager` checks which API keys are configured, queries each provider's API, and checks for a running Ollama server to discover available models:
46
47	```python
48	from video_processor.providers.manager import ProviderManager
49
50	pm = ProviderManager()
51	# Automatically discovers models from all configured providers + Ollama
52	```
53
54	## Routing preferences
55
56	Each task type has a default preference order:
57
58	\| Task \| Preference \|
59	\|------\|-----------\|
60	\| Vision \| Gemini Flash → GPT-4o → Claude Sonnet → Ollama \|
61	\| Chat \| Claude Sonnet → GPT-4o → Gemini Flash → Ollama \|
62	\| Transcription \| Local Whisper → Whisper-1 → Gemini Flash \|
63
64	Ollama acts as the last-resort fallback — if no cloud API keys are set but Ollama is running, it is used automatically.
65
66	## Manual override
67
68	```python
69	pm = ProviderManager(
70	vision_model="gpt-4o",
71	chat_model="claude-sonnet-4-5-20250929",
72	provider="openai", # Force a specific provider
73	)
74
75	# Or use Ollama for fully offline processing
76	pm = ProviderManager(provider="ollama")
77	```
78
79	## BaseProvider interface
80
81	All providers implement:
82

M docs/cli-reference.md

+3 -3

		--- docs/cli-reference.md
		+++ docs/cli-reference.md
		@@ -17,11 +17,11 @@
17	17	\| `--use-gpu` \| FLAG \| off \| Enable GPU acceleration \|
18	18	\| `--sampling-rate` \| FLOAT \| 0.5 \| Frame sampling rate (fps) \|
19	19	\| `--change-threshold` \| FLOAT \| 0.15 \| Visual change threshold \|
20	20	\| `--periodic-capture` \| FLOAT \| 30.0 \| Capture a frame every N seconds regardless of change (0 to disable) \|
21	21	\| `--title` \| TEXT \| auto \| Report title \|
22		-\| `-p`, `--provider` \| `auto\\|openai\\|anthropic\\|gemini` \| `auto` \| API provider \|
	22	+\| `-p`, `--provider` \| `auto\\|openai\\|anthropic\\|gemini\\|ollama` \| `auto` \| API provider \|
23	23	\| `--vision-model` \| TEXT \| auto \| Override vision model \|
24	24	\| `--chat-model` \| TEXT \| auto \| Override chat model \|
25	25
26	26	---
27	27
		@@ -38,11 +38,11 @@
38	38	\| `-i`, `--input-dir` \| PATH \| required \| Directory containing videos \|
39	39	\| `-o`, `--output` \| PATH \| required \| Output directory \|
40	40	\| `--depth` \| `basic\\|standard\\|comprehensive` \| `standard` \| Processing depth \|
41	41	\| `--pattern` \| TEXT \| `.mp4,.mkv,.avi,.mov,*.webm` \| File glob patterns \|
42	42	\| `--title` \| TEXT \| `Batch Processing Results` \| Batch title \|
43		-\| `-p`, `--provider` \| `auto\\|openai\\|anthropic\\|gemini` \| `auto` \| API provider \|
	43	+\| `-p`, `--provider` \| `auto\\|openai\\|anthropic\\|gemini\\|ollama` \| `auto` \| API provider \|
44	44	\| `--vision-model` \| TEXT \| auto \| Override vision model \|
45	45	\| `--chat-model` \| TEXT \| auto \| Override chat model \|
46	46	\| `--source` \| `local\\|gdrive\\|dropbox` \| `local` \| Video source \|
47	47	\| `--folder-id` \| TEXT \| none \| Google Drive folder ID \|
48	48	\| `--folder-path` \| TEXT \| none \| Cloud folder path \|
		@@ -90,11 +90,11 @@
90	90	\|--------\|------\|---------\|-------------\|
91	91	\| `-i`, `--input` \| PATH \| required \| Input video file path \|
92	92	\| `-o`, `--output` \| PATH \| required \| Output directory \|
93	93	\| `--depth` \| `basic\\|standard\\|comprehensive` \| `standard` \| Initial processing depth (agent may adapt) \|
94	94	\| `--title` \| TEXT \| auto \| Report title \|
95		-\| `-p`, `--provider` \| `auto\\|openai\\|anthropic\\|gemini` \| `auto` \| API provider \|
	95	+\| `-p`, `--provider` \| `auto\\|openai\\|anthropic\\|gemini\\|ollama` \| `auto` \| API provider \|
96	96	\| `--vision-model` \| TEXT \| auto \| Override vision model \|
97	97	\| `--chat-model` \| TEXT \| auto \| Override chat model \|
98	98
99	99	---
100	100
101	101

	--- docs/cli-reference.md
	+++ docs/cli-reference.md
	@@ -17,11 +17,11 @@
17	\| `--use-gpu` \| FLAG \| off \| Enable GPU acceleration \|
18	\| `--sampling-rate` \| FLOAT \| 0.5 \| Frame sampling rate (fps) \|
19	\| `--change-threshold` \| FLOAT \| 0.15 \| Visual change threshold \|
20	\| `--periodic-capture` \| FLOAT \| 30.0 \| Capture a frame every N seconds regardless of change (0 to disable) \|
21	\| `--title` \| TEXT \| auto \| Report title \|
22	\| `-p`, `--provider` \| `auto\\|openai\\|anthropic\\|gemini` \| `auto` \| API provider \|
23	\| `--vision-model` \| TEXT \| auto \| Override vision model \|
24	\| `--chat-model` \| TEXT \| auto \| Override chat model \|
25
26	---
27
	@@ -38,11 +38,11 @@
38	\| `-i`, `--input-dir` \| PATH \| required \| Directory containing videos \|
39	\| `-o`, `--output` \| PATH \| required \| Output directory \|
40	\| `--depth` \| `basic\\|standard\\|comprehensive` \| `standard` \| Processing depth \|
41	\| `--pattern` \| TEXT \| `.mp4,.mkv,.avi,.mov,*.webm` \| File glob patterns \|
42	\| `--title` \| TEXT \| `Batch Processing Results` \| Batch title \|
43	\| `-p`, `--provider` \| `auto\\|openai\\|anthropic\\|gemini` \| `auto` \| API provider \|
44	\| `--vision-model` \| TEXT \| auto \| Override vision model \|
45	\| `--chat-model` \| TEXT \| auto \| Override chat model \|
46	\| `--source` \| `local\\|gdrive\\|dropbox` \| `local` \| Video source \|
47	\| `--folder-id` \| TEXT \| none \| Google Drive folder ID \|
48	\| `--folder-path` \| TEXT \| none \| Cloud folder path \|
	@@ -90,11 +90,11 @@
90	\|--------\|------\|---------\|-------------\|
91	\| `-i`, `--input` \| PATH \| required \| Input video file path \|
92	\| `-o`, `--output` \| PATH \| required \| Output directory \|
93	\| `--depth` \| `basic\\|standard\\|comprehensive` \| `standard` \| Initial processing depth (agent may adapt) \|
94	\| `--title` \| TEXT \| auto \| Report title \|
95	\| `-p`, `--provider` \| `auto\\|openai\\|anthropic\\|gemini` \| `auto` \| API provider \|
96	\| `--vision-model` \| TEXT \| auto \| Override vision model \|
97	\| `--chat-model` \| TEXT \| auto \| Override chat model \|
98
99	---
100
101

	--- docs/cli-reference.md
	+++ docs/cli-reference.md
	@@ -17,11 +17,11 @@
17	\| `--use-gpu` \| FLAG \| off \| Enable GPU acceleration \|
18	\| `--sampling-rate` \| FLOAT \| 0.5 \| Frame sampling rate (fps) \|
19	\| `--change-threshold` \| FLOAT \| 0.15 \| Visual change threshold \|
20	\| `--periodic-capture` \| FLOAT \| 30.0 \| Capture a frame every N seconds regardless of change (0 to disable) \|
21	\| `--title` \| TEXT \| auto \| Report title \|
22	\| `-p`, `--provider` \| `auto\\|openai\\|anthropic\\|gemini\\|ollama` \| `auto` \| API provider \|
23	\| `--vision-model` \| TEXT \| auto \| Override vision model \|
24	\| `--chat-model` \| TEXT \| auto \| Override chat model \|
25
26	---
27
	@@ -38,11 +38,11 @@
38	\| `-i`, `--input-dir` \| PATH \| required \| Directory containing videos \|
39	\| `-o`, `--output` \| PATH \| required \| Output directory \|
40	\| `--depth` \| `basic\\|standard\\|comprehensive` \| `standard` \| Processing depth \|
41	\| `--pattern` \| TEXT \| `.mp4,.mkv,.avi,.mov,*.webm` \| File glob patterns \|
42	\| `--title` \| TEXT \| `Batch Processing Results` \| Batch title \|
43	\| `-p`, `--provider` \| `auto\\|openai\\|anthropic\\|gemini\\|ollama` \| `auto` \| API provider \|
44	\| `--vision-model` \| TEXT \| auto \| Override vision model \|
45	\| `--chat-model` \| TEXT \| auto \| Override chat model \|
46	\| `--source` \| `local\\|gdrive\\|dropbox` \| `local` \| Video source \|
47	\| `--folder-id` \| TEXT \| none \| Google Drive folder ID \|
48	\| `--folder-path` \| TEXT \| none \| Cloud folder path \|
	@@ -90,11 +90,11 @@
90	\|--------\|------\|---------\|-------------\|
91	\| `-i`, `--input` \| PATH \| required \| Input video file path \|
92	\| `-o`, `--output` \| PATH \| required \| Output directory \|
93	\| `--depth` \| `basic\\|standard\\|comprehensive` \| `standard` \| Initial processing depth (agent may adapt) \|
94	\| `--title` \| TEXT \| auto \| Report title \|
95	\| `-p`, `--provider` \| `auto\\|openai\\|anthropic\\|gemini\\|ollama` \| `auto` \| API provider \|
96	\| `--vision-model` \| TEXT \| auto \| Override vision model \|
97	\| `--chat-model` \| TEXT \| auto \| Override chat model \|
98
99	---
100
101

M docs/getting-started/configuration.md

+6 -3

		--- docs/getting-started/configuration.md
		+++ docs/getting-started/configuration.md
		@@ -5,22 +5,25 @@
5	5	\| Variable \| Description \|
6	6	\|----------\|-------------\|
7	7	\| `OPENAI_API_KEY` \| OpenAI API key \|
8	8	\| `ANTHROPIC_API_KEY` \| Anthropic API key \|
9	9	\| `GEMINI_API_KEY` \| Google Gemini API key \|
	10	+\| `OLLAMA_HOST` \| Ollama server URL (default: `http://localhost:11434`) \|
10	11	\| `GOOGLE_APPLICATION_CREDENTIALS` \| Path to Google service account JSON (for Drive) \|
11	12	\| `CACHE_DIR` \| Directory for API response caching \|
12	13
13	14	## Provider routing
14	15
15	16	PlanOpticon auto-discovers available models and routes each task to the best option:
16	17
17	18	\| Task \| Default preference \|
18	19	\|------\|--------------------\|
19		-\| Vision (diagrams) \| Gemini Flash > GPT-4o > Claude Sonnet \|
20		-\| Chat (analysis) \| Claude Sonnet > GPT-4o > Gemini Flash \|
21		-\| Transcription \| Whisper-1 > Gemini Flash \|
	20	+\| Vision (diagrams) \| Gemini Flash > GPT-4o > Claude Sonnet > Ollama \|
	21	+\| Chat (analysis) \| Claude Sonnet > GPT-4o > Gemini Flash > Ollama \|
	22	+\| Transcription \| Local Whisper > Whisper-1 > Gemini Flash \|
	23	+
	24	+If no cloud API keys are configured, PlanOpticon automatically falls back to Ollama when a local server is running. This enables fully offline operation when paired with local Whisper for transcription.
22	25
23	26	Override with `--provider`, `--vision-model`, or `--chat-model` flags.
24	27
25	28	## Frame sampling
26	29
27	30

	--- docs/getting-started/configuration.md
	+++ docs/getting-started/configuration.md
	@@ -5,22 +5,25 @@
5	\| Variable \| Description \|
6	\|----------\|-------------\|
7	\| `OPENAI_API_KEY` \| OpenAI API key \|
8	\| `ANTHROPIC_API_KEY` \| Anthropic API key \|
9	\| `GEMINI_API_KEY` \| Google Gemini API key \|

10	\| `GOOGLE_APPLICATION_CREDENTIALS` \| Path to Google service account JSON (for Drive) \|
11	\| `CACHE_DIR` \| Directory for API response caching \|
12
13	## Provider routing
14
15	PlanOpticon auto-discovers available models and routes each task to the best option:
16
17	\| Task \| Default preference \|
18	\|------\|--------------------\|
19	\| Vision (diagrams) \| Gemini Flash > GPT-4o > Claude Sonnet \|
20	\| Chat (analysis) \| Claude Sonnet > GPT-4o > Gemini Flash \|
21	\| Transcription \| Whisper-1 > Gemini Flash \|


22
23	Override with `--provider`, `--vision-model`, or `--chat-model` flags.
24
25	## Frame sampling
26
27

	--- docs/getting-started/configuration.md
	+++ docs/getting-started/configuration.md
	@@ -5,22 +5,25 @@
5	\| Variable \| Description \|
6	\|----------\|-------------\|
7	\| `OPENAI_API_KEY` \| OpenAI API key \|
8	\| `ANTHROPIC_API_KEY` \| Anthropic API key \|
9	\| `GEMINI_API_KEY` \| Google Gemini API key \|
10	\| `OLLAMA_HOST` \| Ollama server URL (default: `http://localhost:11434`) \|
11	\| `GOOGLE_APPLICATION_CREDENTIALS` \| Path to Google service account JSON (for Drive) \|
12	\| `CACHE_DIR` \| Directory for API response caching \|
13
14	## Provider routing
15
16	PlanOpticon auto-discovers available models and routes each task to the best option:
17
18	\| Task \| Default preference \|
19	\|------\|--------------------\|
20	\| Vision (diagrams) \| Gemini Flash > GPT-4o > Claude Sonnet > Ollama \|
21	\| Chat (analysis) \| Claude Sonnet > GPT-4o > Gemini Flash > Ollama \|
22	\| Transcription \| Local Whisper > Whisper-1 > Gemini Flash \|
23
24	If no cloud API keys are configured, PlanOpticon automatically falls back to Ollama when a local server is running. This enables fully offline operation when paired with local Whisper for transcription.
25
26	Override with `--provider`, `--vision-model`, or `--chat-model` flags.
27
28	## Frame sampling
29
30

M docs/getting-started/installation.md

+20 -1

		--- docs/getting-started/installation.md
		+++ docs/getting-started/installation.md
		@@ -62,11 +62,15 @@
62	62
63	63	Download from [ffmpeg.org](https://ffmpeg.org/download.html) and add to PATH.
64	64
65	65	## API keys
66	66
67		-You need at least one AI provider API key. Set them as environment variables:
	67	+You need at least one AI provider API key or a running Ollama server.
	68	+
	69	+### Cloud providers
	70	+
	71	+Set API keys as environment variables:
68	72
69	73	```bash
70	74	export OPENAI_API_KEY="sk-..."
71	75	export ANTHROPIC_API_KEY="sk-ant-..."
72	76	export GEMINI_API_KEY="AI..."
		@@ -78,6 +82,21 @@
78	82	OPENAI_API_KEY=sk-...
79	83	ANTHROPIC_API_KEY=sk-ant-...
80	84	GEMINI_API_KEY=AI...
81	85	```
82	86
	87	+### Ollama (fully offline)
	88	+
	89	+No API keys needed — just install and run [Ollama](https://ollama.com):
	90	+
	91	+```bash
	92	+# Install Ollama, then pull models
	93	+ollama pull llama3.2 # Chat/analysis
	94	+ollama pull llava # Vision (diagram detection)
	95	+
	96	+# Start the server (if not already running)
	97	+ollama serve
	98	+```
	99	+
	100	+PlanOpticon auto-detects Ollama and uses it as a fallback when no cloud API keys are set. For a fully offline pipeline, pair Ollama with local Whisper transcription (`pip install planopticon[gpu]`).
	101	+
83	102	PlanOpticon will automatically discover which providers are available and route to the best model for each task.
84	103

	--- docs/getting-started/installation.md
	+++ docs/getting-started/installation.md
	@@ -62,11 +62,15 @@
62
63	Download from [ffmpeg.org](https://ffmpeg.org/download.html) and add to PATH.
64
65	## API keys
66
67	You need at least one AI provider API key. Set them as environment variables:




68
69	```bash
70	export OPENAI_API_KEY="sk-..."
71	export ANTHROPIC_API_KEY="sk-ant-..."
72	export GEMINI_API_KEY="AI..."
	@@ -78,6 +82,21 @@
78	OPENAI_API_KEY=sk-...
79	ANTHROPIC_API_KEY=sk-ant-...
80	GEMINI_API_KEY=AI...
81	```
82















83	PlanOpticon will automatically discover which providers are available and route to the best model for each task.
84

	--- docs/getting-started/installation.md
	+++ docs/getting-started/installation.md
	@@ -62,11 +62,15 @@
62
63	Download from [ffmpeg.org](https://ffmpeg.org/download.html) and add to PATH.
64
65	## API keys
66
67	You need at least one AI provider API key or a running Ollama server.
68
69	### Cloud providers
70
71	Set API keys as environment variables:
72
73	```bash
74	export OPENAI_API_KEY="sk-..."
75	export ANTHROPIC_API_KEY="sk-ant-..."
76	export GEMINI_API_KEY="AI..."
	@@ -78,6 +82,21 @@
82	OPENAI_API_KEY=sk-...
83	ANTHROPIC_API_KEY=sk-ant-...
84	GEMINI_API_KEY=AI...
85	```
86
87	### Ollama (fully offline)
88
89	No API keys needed — just install and run [Ollama](https://ollama.com):
90
91	```bash
92	# Install Ollama, then pull models
93	ollama pull llama3.2 # Chat/analysis
94	ollama pull llava # Vision (diagram detection)
95
96	# Start the server (if not already running)
97	ollama serve
98	```
99
100	PlanOpticon auto-detects Ollama and uses it as a fallback when no cloud API keys are set. For a fully offline pipeline, pair Ollama with local Whisper transcription (`pip install planopticon[gpu]`).
101
102	PlanOpticon will automatically discover which providers are available and route to the best model for each task.
103

M docs/getting-started/quickstart.md

		--- docs/getting-started/quickstart.md
		+++ docs/getting-started/quickstart.md
		@@ -35,10 +35,13 @@
35	35	# Auto-detect best available (default)
36	36	planopticon analyze -i video.mp4 -o ./out
37	37
38	38	# Force a specific provider
39	39	planopticon analyze -i video.mp4 -o ./out --provider openai
	40	+
	41	+# Use Ollama for fully offline processing (no API keys needed)
	42	+planopticon analyze -i video.mp4 -o ./out --provider ollama
40	43
41	44	# Override specific models
42	45	planopticon analyze -i video.mp4 -o ./out \
43	46	--vision-model gpt-4o \
44	47	--chat-model claude-sonnet-4-5-20250929
45	48

	--- docs/getting-started/quickstart.md
	+++ docs/getting-started/quickstart.md
	@@ -35,10 +35,13 @@
35	# Auto-detect best available (default)
36	planopticon analyze -i video.mp4 -o ./out
37
38	# Force a specific provider
39	planopticon analyze -i video.mp4 -o ./out --provider openai



40
41	# Override specific models
42	planopticon analyze -i video.mp4 -o ./out \
43	--vision-model gpt-4o \
44	--chat-model claude-sonnet-4-5-20250929
45

	--- docs/getting-started/quickstart.md
	+++ docs/getting-started/quickstart.md
	@@ -35,10 +35,13 @@
35	# Auto-detect best available (default)
36	planopticon analyze -i video.mp4 -o ./out
37
38	# Force a specific provider
39	planopticon analyze -i video.mp4 -o ./out --provider openai
40
41	# Use Ollama for fully offline processing (no API keys needed)
42	planopticon analyze -i video.mp4 -o ./out --provider ollama
43
44	# Override specific models
45	planopticon analyze -i video.mp4 -o ./out \
46	--vision-model gpt-4o \
47	--chat-model claude-sonnet-4-5-20250929
48

M tests/test_providers.py

+91 -3

		--- tests/test_providers.py
		+++ tests/test_providers.py
		@@ -1,8 +1,10 @@
1	1	"""Tests for the provider abstraction layer."""
2	2
3	3	from unittest.mock import MagicMock, patch
	4	+
	5	+import pytest
4	6
5	7	from video_processor.providers.base import BaseProvider, ModelInfo
6	8	from video_processor.providers.manager import ProviderManager
7	9
8	10
		@@ -107,21 +109,27 @@
107	109	assert "anthropic/claude-sonnet-4-5-20250929" == used["chat"]
108	110
109	111
110	112	class TestDiscovery:
111	113	@patch("video_processor.providers.discovery._cached_models", None)
	114	+ @patch(
	115	+ "video_processor.providers.ollama_provider.OllamaProvider.is_available", return_value=False
	116	+ )
112	117	@patch.dict("os.environ", {}, clear=True)
113		- def test_discover_skips_missing_keys(self):
	118	+ def test_discover_skips_missing_keys(self, mock_ollama):
114	119	from video_processor.providers.discovery import discover_available_models
115	120
116		- # No API keys -> empty list, no errors
	121	+ # No API keys and no Ollama -> empty list, no errors
117	122	models = discover_available_models(api_keys={"openai": "", "anthropic": "", "gemini": ""})
118	123	assert models == []
119	124
120	125	@patch.dict("os.environ", {}, clear=True)
	126	+ @patch(
	127	+ "video_processor.providers.ollama_provider.OllamaProvider.is_available", return_value=False
	128	+ )
121	129	@patch("video_processor.providers.discovery._cached_models", None)
122		- def test_discover_caches_results(self):
	130	+ def test_discover_caches_results(self, mock_ollama):
123	131	from video_processor.providers import discovery
124	132
125	133	models = discovery.discover_available_models(
126	134	api_keys={"openai": "", "anthropic": "", "gemini": ""}
127	135	)
		@@ -131,5 +139,85 @@
131	139	assert models2 == [] # Still cached empty result
132	140
133	141	# Force refresh
134	142	discovery.clear_discovery_cache()
135	143	# Would try to connect with real key, so skip that test
	144	+
	145	+
	146	+class TestOllamaProvider:
	147	+ @patch("video_processor.providers.ollama_provider.requests")
	148	+ def test_is_available_when_running(self, mock_requests):
	149	+ mock_resp = MagicMock()
	150	+ mock_resp.status_code = 200
	151	+ mock_requests.get.return_value = mock_resp
	152	+
	153	+ from video_processor.providers.ollama_provider import OllamaProvider
	154	+
	155	+ assert OllamaProvider.is_available()
	156	+
	157	+ @patch("video_processor.providers.ollama_provider.requests")
	158	+ def test_is_available_when_not_running(self, mock_requests):
	159	+ mock_requests.get.side_effect = ConnectionError
	160	+
	161	+ from video_processor.providers.ollama_provider import OllamaProvider
	162	+
	163	+ assert not OllamaProvider.is_available()
	164	+
	165	+ @patch("video_processor.providers.ollama_provider.requests")
	166	+ @patch("video_processor.providers.ollama_provider.OpenAI")
	167	+ def test_transcribe_raises(self, mock_openai, mock_requests):
	168	+ from video_processor.providers.ollama_provider import OllamaProvider
	169	+
	170	+ provider = OllamaProvider()
	171	+ with pytest.raises(NotImplementedError):
	172	+ provider.transcribe_audio("/tmp/test.wav")
	173	+
	174	+ @patch("video_processor.providers.ollama_provider.requests")
	175	+ @patch("video_processor.providers.ollama_provider.OpenAI")
	176	+ def test_list_models(self, mock_openai, mock_requests):
	177	+ mock_resp = MagicMock()
	178	+ mock_resp.status_code = 200
	179	+ mock_resp.json.return_value = {
	180	+ "models": [
	181	+ {"name": "llama3.2:latest", "details": {"family": "llama"}},
	182	+ {"name": "llava:13b", "details": {"family": "llava"}},
	183	+ ]
	184	+ }
	185	+ mock_requests.get.return_value = mock_resp
	186	+
	187	+ from video_processor.providers.ollama_provider import OllamaProvider
	188	+
	189	+ provider = OllamaProvider()
	190	+ models = provider.list_models()
	191	+ assert len(models) == 2
	192	+ assert models[0].provider == "ollama"
	193	+
	194	+ # llava should have vision capability
	195	+ llava = [m for m in models if "llava" in m.id][0]
	196	+ assert "vision" in llava.capabilities
	197	+
	198	+ # llama should have only chat
	199	+ llama = [m for m in models if "llama" in m.id][0]
	200	+ assert "chat" in llama.capabilities
	201	+ assert "vision" not in llama.capabilities
	202	+
	203	+ def test_provider_for_model_ollama_via_discovery(self):
	204	+ mgr = ProviderManager()
	205	+ mgr._available_models = [
	206	+ ModelInfo(id="llama3.2:latest", provider="ollama", capabilities=["chat"]),
	207	+ ]
	208	+ assert mgr._provider_for_model("llama3.2:latest") == "ollama"
	209	+
	210	+ def test_provider_for_model_ollama_fuzzy_tag(self):
	211	+ mgr = ProviderManager()
	212	+ mgr._available_models = [
	213	+ ModelInfo(id="llama3.2:latest", provider="ollama", capabilities=["chat"]),
	214	+ ]
	215	+ # Should match "llama3.2" to "llama3.2:latest" via prefix
	216	+ assert mgr._provider_for_model("llama3.2") == "ollama"
	217	+
	218	+ def test_init_forced_provider_ollama(self):
	219	+ mgr = ProviderManager(provider="ollama")
	220	+ # Ollama defaults are empty (resolved dynamically)
	221	+ assert mgr.vision_model == ""
	222	+ assert mgr.chat_model == ""
	223	+ assert mgr.transcription_model == ""
136	224

	--- tests/test_providers.py
	+++ tests/test_providers.py
	@@ -1,8 +1,10 @@
1	"""Tests for the provider abstraction layer."""
2
3	from unittest.mock import MagicMock, patch


4
5	from video_processor.providers.base import BaseProvider, ModelInfo
6	from video_processor.providers.manager import ProviderManager
7
8
	@@ -107,21 +109,27 @@
107	assert "anthropic/claude-sonnet-4-5-20250929" == used["chat"]
108
109
110	class TestDiscovery:
111	@patch("video_processor.providers.discovery._cached_models", None)



112	@patch.dict("os.environ", {}, clear=True)
113	def test_discover_skips_missing_keys(self):
114	from video_processor.providers.discovery import discover_available_models
115
116	# No API keys -> empty list, no errors
117	models = discover_available_models(api_keys={"openai": "", "anthropic": "", "gemini": ""})
118	assert models == []
119
120	@patch.dict("os.environ", {}, clear=True)



121	@patch("video_processor.providers.discovery._cached_models", None)
122	def test_discover_caches_results(self):
123	from video_processor.providers import discovery
124
125	models = discovery.discover_available_models(
126	api_keys={"openai": "", "anthropic": "", "gemini": ""}
127	)
	@@ -131,5 +139,85 @@
131	assert models2 == [] # Still cached empty result
132
133	# Force refresh
134	discovery.clear_discovery_cache()
135	# Would try to connect with real key, so skip that test
















































































136

	--- tests/test_providers.py
	+++ tests/test_providers.py
	@@ -1,8 +1,10 @@
1	"""Tests for the provider abstraction layer."""
2
3	from unittest.mock import MagicMock, patch
4
5	import pytest
6
7	from video_processor.providers.base import BaseProvider, ModelInfo
8	from video_processor.providers.manager import ProviderManager
9
10
	@@ -107,21 +109,27 @@
109	assert "anthropic/claude-sonnet-4-5-20250929" == used["chat"]
110
111
112	class TestDiscovery:
113	@patch("video_processor.providers.discovery._cached_models", None)
114	@patch(
115	"video_processor.providers.ollama_provider.OllamaProvider.is_available", return_value=False
116	)
117	@patch.dict("os.environ", {}, clear=True)
118	def test_discover_skips_missing_keys(self, mock_ollama):
119	from video_processor.providers.discovery import discover_available_models
120
121	# No API keys and no Ollama -> empty list, no errors
122	models = discover_available_models(api_keys={"openai": "", "anthropic": "", "gemini": ""})
123	assert models == []
124
125	@patch.dict("os.environ", {}, clear=True)
126	@patch(
127	"video_processor.providers.ollama_provider.OllamaProvider.is_available", return_value=False
128	)
129	@patch("video_processor.providers.discovery._cached_models", None)
130	def test_discover_caches_results(self, mock_ollama):
131	from video_processor.providers import discovery
132
133	models = discovery.discover_available_models(
134	api_keys={"openai": "", "anthropic": "", "gemini": ""}
135	)
	@@ -131,5 +139,85 @@
139	assert models2 == [] # Still cached empty result
140
141	# Force refresh
142	discovery.clear_discovery_cache()
143	# Would try to connect with real key, so skip that test
144
145
146	class TestOllamaProvider:
147	@patch("video_processor.providers.ollama_provider.requests")
148	def test_is_available_when_running(self, mock_requests):
149	mock_resp = MagicMock()
150	mock_resp.status_code = 200
151	mock_requests.get.return_value = mock_resp
152
153	from video_processor.providers.ollama_provider import OllamaProvider
154
155	assert OllamaProvider.is_available()
156
157	@patch("video_processor.providers.ollama_provider.requests")
158	def test_is_available_when_not_running(self, mock_requests):
159	mock_requests.get.side_effect = ConnectionError
160
161	from video_processor.providers.ollama_provider import OllamaProvider
162
163	assert not OllamaProvider.is_available()
164
165	@patch("video_processor.providers.ollama_provider.requests")
166	@patch("video_processor.providers.ollama_provider.OpenAI")
167	def test_transcribe_raises(self, mock_openai, mock_requests):
168	from video_processor.providers.ollama_provider import OllamaProvider
169
170	provider = OllamaProvider()
171	with pytest.raises(NotImplementedError):
172	provider.transcribe_audio("/tmp/test.wav")
173
174	@patch("video_processor.providers.ollama_provider.requests")
175	@patch("video_processor.providers.ollama_provider.OpenAI")
176	def test_list_models(self, mock_openai, mock_requests):
177	mock_resp = MagicMock()
178	mock_resp.status_code = 200
179	mock_resp.json.return_value = {
180	"models": [
181	{"name": "llama3.2:latest", "details": {"family": "llama"}},
182	{"name": "llava:13b", "details": {"family": "llava"}},
183	]
184	}
185	mock_requests.get.return_value = mock_resp
186
187	from video_processor.providers.ollama_provider import OllamaProvider
188
189	provider = OllamaProvider()
190	models = provider.list_models()
191	assert len(models) == 2
192	assert models[0].provider == "ollama"
193
194	# llava should have vision capability
195	llava = [m for m in models if "llava" in m.id][0]
196	assert "vision" in llava.capabilities
197
198	# llama should have only chat
199	llama = [m for m in models if "llama" in m.id][0]
200	assert "chat" in llama.capabilities
201	assert "vision" not in llama.capabilities
202
203	def test_provider_for_model_ollama_via_discovery(self):
204	mgr = ProviderManager()
205	mgr._available_models = [
206	ModelInfo(id="llama3.2:latest", provider="ollama", capabilities=["chat"]),
207	]
208	assert mgr._provider_for_model("llama3.2:latest") == "ollama"
209
210	def test_provider_for_model_ollama_fuzzy_tag(self):
211	mgr = ProviderManager()
212	mgr._available_models = [
213	ModelInfo(id="llama3.2:latest", provider="ollama", capabilities=["chat"]),
214	]
215	# Should match "llama3.2" to "llama3.2:latest" via prefix
216	assert mgr._provider_for_model("llama3.2") == "ollama"
217
218	def test_init_forced_provider_ollama(self):
219	mgr = ProviderManager(provider="ollama")
220	# Ollama defaults are empty (resolved dynamically)
221	assert mgr.vision_model == ""
222	assert mgr.chat_model == ""
223	assert mgr.transcription_model == ""
224

M video_processor/cli/commands.py

+9 -7

		--- video_processor/cli/commands.py
		+++ video_processor/cli/commands.py
		@@ -73,11 +73,11 @@
73	73	)
74	74	@click.option("--title", type=str, help="Title for the analysis report")
75	75	@click.option(
76	76	"--provider",
77	77	"-p",
78		- type=click.Choice(["auto", "openai", "anthropic", "gemini"]),
	78	+ type=click.Choice(["auto", "openai", "anthropic", "gemini", "ollama"]),
79	79	default="auto",
80	80	help="API provider",
81	81	)
82	82	@click.option("--vision-model", type=str, default=None, help="Override model for vision tasks")
83	83	@click.option("--chat-model", type=str, default=None, help="Override model for LLM/chat tasks")
		@@ -154,11 +154,11 @@
154	154	)
155	155	@click.option("--title", type=str, default="Batch Processing Results", help="Batch title")
156	156	@click.option(
157	157	"--provider",
158	158	"-p",
159		- type=click.Choice(["auto", "openai", "anthropic", "gemini"]),
	159	+ type=click.Choice(["auto", "openai", "anthropic", "gemini", "ollama"]),
160	160	default="auto",
161	161	help="API provider",
162	162	)
163	163	@click.option("--vision-model", type=str, default=None, help="Override model for vision tasks")
164	164	@click.option("--chat-model", type=str, default=None, help="Override model for LLM/chat tasks")
		@@ -343,12 +343,14 @@
343	343	"""Discover and display available models from all configured providers."""
344	344	from video_processor.providers.discovery import discover_available_models
345	345
346	346	models = discover_available_models(force_refresh=True)
347	347	if not models:
348		- click.echo("No models discovered. Check that at least one API key is set:")
349		- click.echo(" OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY")
	348	+ click.echo(
	349	+ "No models discovered. Check that at least one API key is set or Ollama is running:"
	350	+ )
	351	+ click.echo(" OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, or `ollama serve`")
350	352	return
351	353
352	354	by_provider: dict[str, list] = {}
353	355	for m in models:
354	356	by_provider.setdefault(m.provider, []).append(m)
		@@ -417,11 +419,11 @@
417	419	)
418	420	@click.option("--title", type=str, help="Title for the analysis report")
419	421	@click.option(
420	422	"--provider",
421	423	"-p",
422		- type=click.Choice(["auto", "openai", "anthropic", "gemini"]),
	424	+ type=click.Choice(["auto", "openai", "anthropic", "gemini", "ollama"]),
423	425	default="auto",
424	426	help="API provider",
425	427	)
426	428	@click.option("--vision-model", type=str, default=None, help="Override model for vision tasks")
427	429	@click.option("--chat-model", type=str, default=None, help="Override model for LLM/chat tasks")
		@@ -511,11 +513,11 @@
511	513	type=click.Choice(["basic", "standard", "comprehensive"]),
512	514	default="standard",
513	515	)
514	516	provider = click.prompt(
515	517	" Provider",
516		- type=click.Choice(["auto", "openai", "anthropic", "gemini"]),
	518	+ type=click.Choice(["auto", "openai", "anthropic", "gemini", "ollama"]),
517	519	default="auto",
518	520	)
519	521	ctx.invoke(
520	522	analyze,
521	523	input=input_path,
		@@ -540,11 +542,11 @@
540	542	type=click.Choice(["basic", "standard", "comprehensive"]),
541	543	default="standard",
542	544	)
543	545	provider = click.prompt(
544	546	" Provider",
545		- type=click.Choice(["auto", "openai", "anthropic", "gemini"]),
	547	+ type=click.Choice(["auto", "openai", "anthropic", "gemini", "ollama"]),
546	548	default="auto",
547	549	)
548	550	ctx.invoke(
549	551	batch,
550	552	input_dir=input_dir,
551	553

	--- video_processor/cli/commands.py
	+++ video_processor/cli/commands.py
	@@ -73,11 +73,11 @@
73	)
74	@click.option("--title", type=str, help="Title for the analysis report")
75	@click.option(
76	"--provider",
77	"-p",
78	type=click.Choice(["auto", "openai", "anthropic", "gemini"]),
79	default="auto",
80	help="API provider",
81	)
82	@click.option("--vision-model", type=str, default=None, help="Override model for vision tasks")
83	@click.option("--chat-model", type=str, default=None, help="Override model for LLM/chat tasks")
	@@ -154,11 +154,11 @@
154	)
155	@click.option("--title", type=str, default="Batch Processing Results", help="Batch title")
156	@click.option(
157	"--provider",
158	"-p",
159	type=click.Choice(["auto", "openai", "anthropic", "gemini"]),
160	default="auto",
161	help="API provider",
162	)
163	@click.option("--vision-model", type=str, default=None, help="Override model for vision tasks")
164	@click.option("--chat-model", type=str, default=None, help="Override model for LLM/chat tasks")
	@@ -343,12 +343,14 @@
343	"""Discover and display available models from all configured providers."""
344	from video_processor.providers.discovery import discover_available_models
345
346	models = discover_available_models(force_refresh=True)
347	if not models:
348	click.echo("No models discovered. Check that at least one API key is set:")
349	click.echo(" OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY")


350	return
351
352	by_provider: dict[str, list] = {}
353	for m in models:
354	by_provider.setdefault(m.provider, []).append(m)
	@@ -417,11 +419,11 @@
417	)
418	@click.option("--title", type=str, help="Title for the analysis report")
419	@click.option(
420	"--provider",
421	"-p",
422	type=click.Choice(["auto", "openai", "anthropic", "gemini"]),
423	default="auto",
424	help="API provider",
425	)
426	@click.option("--vision-model", type=str, default=None, help="Override model for vision tasks")
427	@click.option("--chat-model", type=str, default=None, help="Override model for LLM/chat tasks")
	@@ -511,11 +513,11 @@
511	type=click.Choice(["basic", "standard", "comprehensive"]),
512	default="standard",
513	)
514	provider = click.prompt(
515	" Provider",
516	type=click.Choice(["auto", "openai", "anthropic", "gemini"]),
517	default="auto",
518	)
519	ctx.invoke(
520	analyze,
521	input=input_path,
	@@ -540,11 +542,11 @@
540	type=click.Choice(["basic", "standard", "comprehensive"]),
541	default="standard",
542	)
543	provider = click.prompt(
544	" Provider",
545	type=click.Choice(["auto", "openai", "anthropic", "gemini"]),
546	default="auto",
547	)
548	ctx.invoke(
549	batch,
550	input_dir=input_dir,
551

	--- video_processor/cli/commands.py
	+++ video_processor/cli/commands.py
	@@ -73,11 +73,11 @@
73	)
74	@click.option("--title", type=str, help="Title for the analysis report")
75	@click.option(
76	"--provider",
77	"-p",
78	type=click.Choice(["auto", "openai", "anthropic", "gemini", "ollama"]),
79	default="auto",
80	help="API provider",
81	)
82	@click.option("--vision-model", type=str, default=None, help="Override model for vision tasks")
83	@click.option("--chat-model", type=str, default=None, help="Override model for LLM/chat tasks")
	@@ -154,11 +154,11 @@
154	)
155	@click.option("--title", type=str, default="Batch Processing Results", help="Batch title")
156	@click.option(
157	"--provider",
158	"-p",
159	type=click.Choice(["auto", "openai", "anthropic", "gemini", "ollama"]),
160	default="auto",
161	help="API provider",
162	)
163	@click.option("--vision-model", type=str, default=None, help="Override model for vision tasks")
164	@click.option("--chat-model", type=str, default=None, help="Override model for LLM/chat tasks")
	@@ -343,12 +343,14 @@
343	"""Discover and display available models from all configured providers."""
344	from video_processor.providers.discovery import discover_available_models
345
346	models = discover_available_models(force_refresh=True)
347	if not models:
348	click.echo(
349	"No models discovered. Check that at least one API key is set or Ollama is running:"
350	)
351	click.echo(" OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, or `ollama serve`")
352	return
353
354	by_provider: dict[str, list] = {}
355	for m in models:
356	by_provider.setdefault(m.provider, []).append(m)
	@@ -417,11 +419,11 @@
419	)
420	@click.option("--title", type=str, help="Title for the analysis report")
421	@click.option(
422	"--provider",
423	"-p",
424	type=click.Choice(["auto", "openai", "anthropic", "gemini", "ollama"]),
425	default="auto",
426	help="API provider",
427	)
428	@click.option("--vision-model", type=str, default=None, help="Override model for vision tasks")
429	@click.option("--chat-model", type=str, default=None, help="Override model for LLM/chat tasks")
	@@ -511,11 +513,11 @@
513	type=click.Choice(["basic", "standard", "comprehensive"]),
514	default="standard",
515	)
516	provider = click.prompt(
517	" Provider",
518	type=click.Choice(["auto", "openai", "anthropic", "gemini", "ollama"]),
519	default="auto",
520	)
521	ctx.invoke(
522	analyze,
523	input=input_path,
	@@ -540,11 +542,11 @@
542	type=click.Choice(["basic", "standard", "comprehensive"]),
543	default="standard",
544	)
545	provider = click.prompt(
546	" Provider",
547	type=click.Choice(["auto", "openai", "anthropic", "gemini", "ollama"]),
548	default="auto",
549	)
550	ctx.invoke(
551	batch,
552	input_dir=input_dir,
553

M video_processor/providers/discovery.py

+12

		--- video_processor/providers/discovery.py
		+++ video_processor/providers/discovery.py
		@@ -75,10 +75,22 @@
75	75	logger.info(f"Discovered {len(models)} Gemini models")
76	76	all_models.extend(models)
77	77	except Exception as e:
78	78	logger.warning(f"Gemini discovery failed: {e}")
79	79
	80	+ # Ollama (local, no API key needed)
	81	+ try:
	82	+ from video_processor.providers.ollama_provider import OllamaProvider
	83	+
	84	+ if OllamaProvider.is_available():
	85	+ provider = OllamaProvider()
	86	+ models = provider.list_models()
	87	+ logger.info(f"Discovered {len(models)} Ollama models")
	88	+ all_models.extend(models)
	89	+ except Exception as e:
	90	+ logger.info(f"Ollama discovery skipped: {e}")
	91	+
80	92	# Sort by provider then id
81	93	all_models.sort(key=lambda m: (m.provider, m.id))
82	94	_cached_models = all_models
83	95	logger.info(f"Total discovered models: {len(all_models)}")
84	96	return all_models
85	97

	--- video_processor/providers/discovery.py
	+++ video_processor/providers/discovery.py
	@@ -75,10 +75,22 @@
75	logger.info(f"Discovered {len(models)} Gemini models")
76	all_models.extend(models)
77	except Exception as e:
78	logger.warning(f"Gemini discovery failed: {e}")
79












80	# Sort by provider then id
81	all_models.sort(key=lambda m: (m.provider, m.id))
82	_cached_models = all_models
83	logger.info(f"Total discovered models: {len(all_models)}")
84	return all_models
85

	--- video_processor/providers/discovery.py
	+++ video_processor/providers/discovery.py
	@@ -75,10 +75,22 @@
75	logger.info(f"Discovered {len(models)} Gemini models")
76	all_models.extend(models)
77	except Exception as e:
78	logger.warning(f"Gemini discovery failed: {e}")
79
80	# Ollama (local, no API key needed)
81	try:
82	from video_processor.providers.ollama_provider import OllamaProvider
83
84	if OllamaProvider.is_available():
85	provider = OllamaProvider()
86	models = provider.list_models()
87	logger.info(f"Discovered {len(models)} Ollama models")
88	all_models.extend(models)
89	except Exception as e:
90	logger.info(f"Ollama discovery skipped: {e}")
91
92	# Sort by provider then id
93	all_models.sort(key=lambda m: (m.provider, m.id))
94	_cached_models = all_models
95	logger.info(f"Total discovered models: {len(all_models)}")
96	return all_models
97

M video_processor/providers/manager.py

+27 -2

		--- video_processor/providers/manager.py
		+++ video_processor/providers/manager.py
		@@ -91,10 +91,15 @@
91	91	"gemini": {
92	92	"chat": "gemini-2.5-flash",
93	93	"vision": "gemini-2.5-flash",
94	94	"audio": "gemini-2.5-flash",
95	95	},
	96	+ "ollama": {
	97	+ "chat": "",
	98	+ "vision": "",
	99	+ "audio": "",
	100	+ },
96	101	}
97	102	return defaults.get(provider, {}).get(capability, "")
98	103
99	104	def _get_provider(self, provider_name: str) -> BaseProvider:
100	105	"""Lazily initialize and cache a provider instance."""
		@@ -109,10 +114,14 @@
109	114	self._providers[provider_name] = AnthropicProvider()
110	115	elif provider_name == "gemini":
111	116	from video_processor.providers.gemini_provider import GeminiProvider
112	117
113	118	self._providers[provider_name] = GeminiProvider()
	119	+ elif provider_name == "ollama":
	120	+ from video_processor.providers.ollama_provider import OllamaProvider
	121	+
	122	+ self._providers[provider_name] = OllamaProvider()
114	123	else:
115	124	raise ValueError(f"Unknown provider: {provider_name}")
116	125	return self._providers[provider_name]
117	126
118	127	def _provider_for_model(self, model_id: str) -> str:
		@@ -127,15 +136,18 @@
127	136	return "openai"
128	137	if model_id.startswith("claude-"):
129	138	return "anthropic"
130	139	if model_id.startswith("gemini-"):
131	140	return "gemini"
132		- # Try discovery
	141	+ # Try discovery (exact match, then prefix match for ollama name:tag format)
133	142	models = self._get_available_models()
134	143	for m in models:
135	144	if m.id == model_id:
136	145	return m.provider
	146	+ for m in models:
	147	+ if m.id.startswith(model_id + ":"):
	148	+ return m.provider
137	149	raise ValueError(f"Cannot determine provider for model: {model_id}")
138	150
139	151	def _get_available_models(self) -> list[ModelInfo]:
140	152	if self._available_models is None:
141	153	self._available_models = discover_available_models()
		@@ -159,14 +171,27 @@
159	171	try:
160	172	self._get_provider(prov)
161	173	return prov, model
162	174	except (ValueError, ImportError):
163	175	continue
	176	+
	177	+ # Fallback: try Ollama if available (no API key needed)
	178	+ try:
	179	+ from video_processor.providers.ollama_provider import OllamaProvider
	180	+
	181	+ if OllamaProvider.is_available():
	182	+ provider = self._get_provider("ollama")
	183	+ models = provider.list_models()
	184	+ for m in models:
	185	+ if capability in m.capabilities:
	186	+ return "ollama", m.id
	187	+ except Exception:
	188	+ pass
164	189
165	190	raise RuntimeError(
166	191	f"No provider available for capability '{capability}'. "
167		- "Set an API key for at least one provider."
	192	+ "Set an API key for at least one provider, or start Ollama."
168	193	)
169	194
170	195	def _track(self, provider: BaseProvider, prov_name: str, model: str) -> None:
171	196	"""Record usage from the last API call on a provider."""
172	197	last = getattr(provider, "_last_usage", None)
173	198
174	199	ADDED video_processor/providers/ollama_provider.py

	--- video_processor/providers/manager.py
	+++ video_processor/providers/manager.py
	@@ -91,10 +91,15 @@
91	"gemini": {
92	"chat": "gemini-2.5-flash",
93	"vision": "gemini-2.5-flash",
94	"audio": "gemini-2.5-flash",
95	},





96	}
97	return defaults.get(provider, {}).get(capability, "")
98
99	def _get_provider(self, provider_name: str) -> BaseProvider:
100	"""Lazily initialize and cache a provider instance."""
	@@ -109,10 +114,14 @@
109	self._providers[provider_name] = AnthropicProvider()
110	elif provider_name == "gemini":
111	from video_processor.providers.gemini_provider import GeminiProvider
112
113	self._providers[provider_name] = GeminiProvider()




114	else:
115	raise ValueError(f"Unknown provider: {provider_name}")
116	return self._providers[provider_name]
117
118	def _provider_for_model(self, model_id: str) -> str:
	@@ -127,15 +136,18 @@
127	return "openai"
128	if model_id.startswith("claude-"):
129	return "anthropic"
130	if model_id.startswith("gemini-"):
131	return "gemini"
132	# Try discovery
133	models = self._get_available_models()
134	for m in models:
135	if m.id == model_id:
136	return m.provider



137	raise ValueError(f"Cannot determine provider for model: {model_id}")
138
139	def _get_available_models(self) -> list[ModelInfo]:
140	if self._available_models is None:
141	self._available_models = discover_available_models()
	@@ -159,14 +171,27 @@
159	try:
160	self._get_provider(prov)
161	return prov, model
162	except (ValueError, ImportError):
163	continue













164
165	raise RuntimeError(
166	f"No provider available for capability '{capability}'. "
167	"Set an API key for at least one provider."
168	)
169
170	def _track(self, provider: BaseProvider, prov_name: str, model: str) -> None:
171	"""Record usage from the last API call on a provider."""
172	last = getattr(provider, "_last_usage", None)
173
174	DDED video_processor/providers/ollama_provider.py

	--- video_processor/providers/manager.py
	+++ video_processor/providers/manager.py
	@@ -91,10 +91,15 @@
91	"gemini": {
92	"chat": "gemini-2.5-flash",
93	"vision": "gemini-2.5-flash",
94	"audio": "gemini-2.5-flash",
95	},
96	"ollama": {
97	"chat": "",
98	"vision": "",
99	"audio": "",
100	},
101	}
102	return defaults.get(provider, {}).get(capability, "")
103
104	def _get_provider(self, provider_name: str) -> BaseProvider:
105	"""Lazily initialize and cache a provider instance."""
	@@ -109,10 +114,14 @@
114	self._providers[provider_name] = AnthropicProvider()
115	elif provider_name == "gemini":
116	from video_processor.providers.gemini_provider import GeminiProvider
117
118	self._providers[provider_name] = GeminiProvider()
119	elif provider_name == "ollama":
120	from video_processor.providers.ollama_provider import OllamaProvider
121
122	self._providers[provider_name] = OllamaProvider()
123	else:
124	raise ValueError(f"Unknown provider: {provider_name}")
125	return self._providers[provider_name]
126
127	def _provider_for_model(self, model_id: str) -> str:
	@@ -127,15 +136,18 @@
136	return "openai"
137	if model_id.startswith("claude-"):
138	return "anthropic"
139	if model_id.startswith("gemini-"):
140	return "gemini"
141	# Try discovery (exact match, then prefix match for ollama name:tag format)
142	models = self._get_available_models()
143	for m in models:
144	if m.id == model_id:
145	return m.provider
146	for m in models:
147	if m.id.startswith(model_id + ":"):
148	return m.provider
149	raise ValueError(f"Cannot determine provider for model: {model_id}")
150
151	def _get_available_models(self) -> list[ModelInfo]:
152	if self._available_models is None:
153	self._available_models = discover_available_models()
	@@ -159,14 +171,27 @@
171	try:
172	self._get_provider(prov)
173	return prov, model
174	except (ValueError, ImportError):
175	continue
176
177	# Fallback: try Ollama if available (no API key needed)
178	try:
179	from video_processor.providers.ollama_provider import OllamaProvider
180
181	if OllamaProvider.is_available():
182	provider = self._get_provider("ollama")
183	models = provider.list_models()
184	for m in models:
185	if capability in m.capabilities:
186	return "ollama", m.id
187	except Exception:
188	pass
189
190	raise RuntimeError(
191	f"No provider available for capability '{capability}'. "
192	"Set an API key for at least one provider, or start Ollama."
193	)
194
195	def _track(self, provider: BaseProvider, prov_name: str, model: str) -> None:
196	"""Record usage from the last API call on a provider."""
197	last = getattr(provider, "_last_usage", None)
198
199	DDED video_processor/providers/ollama_provider.py

A video_processor/providers/ollama_provider.py

+109

		--- a/video_processor/providers/ollama_provider.py
		+++ b/video_processor/providers/ollama_provider.py
		@@ -0,0 +1,109 @@
	1	+"""Ollama provider implementation using OpenAI-compatible API."""
	2	+
	3	+import base64
	4	+import logging
	5	+import os
	6	+from pathlib import Path
	7	+from typing import Optional
	8	+
	9	+import requests
	10	+from openai import OpenAI
	11	+
	12	+from video_processor.providers.base import BaseProvider, ModelInfo
	13	+
	14	+logger = logging.getLogger(__name__)
	15	+
	16	+# Known vision-capable model families (base name before the colon/tag)
	17	+_VISION_FAMILIES = {
	18	+ "llava",
	19	+ "llava-llama3",
	20	+ "llava-phi3",
	21	+ "llama3.2-vision",
	22	+ "moondream",
	23	+ "bakllava",
	24	+ "minicpm-v",
	25	+ "deepseek-vl",
	26	+ "internvl2",
	27	+}
	28	+
	29	+DEFAULT_HOST = "http://localhost:11434"
	30	+
	31	+
	32	+class OllamaProvider(BaseProvider):
	33	+ """Ollama local LLM provider via OpenAI-compatible API."""
	34	+
	35	+ provider_name = model = model or self._dprompt_tokens", 0) self.host ULT_HOST)
	36	+ model = model or secmethod
	37	+ def is_available(host: Optional[str] = None) -> bool:
	38	+ """Check if an Ollama server is running and reachable."""
	39	+ host = host or os.getenv("OLLAMA_HOST", DEFAULT_HOST)
	40	+ try:
	41	+ resp = requests.get(f"{host}/api/tags", timeout=3)
	42	+ return resp.status_code == 200
	43	+ except Exception:
	44	+ return False
	45	+
	46	+ @property
	47	+ def _default_model(self) -> str:
	48	+ models = self._get_models()
	49	+ for m in models:
	50	+ if "chat" in m.capabilities:
	51	+ return m.id
	52	+ return "llama3.2:latest"
	53	+
	54	+ @property
	55	+ def _default_vision_model(self) -> Optional[str]:
	56	+ models = self._get_models()
	57	+ for m in models:
	58	+ if "vision" in m.capabilities:
	59	+ return m.id
	60	+ return None
	61	+
	62	+ def _get_models(self) -> list[ModelInfo]:
	63	+ if self._models_cache is None:
	64	+ self._models_cache = self.list_models()
	65	+ return self._models_cache
	66	+
	67	+ def chat(
	68	+ self,
	69	+ messages: list[ model = model or self._dprompt_tokens", 0) self.host ULT_HOST)
	70	+ model = model or setions.create(
	71	+ model=model,
	72	+ messages=messages,
	73	+ max_tokens=max_tokens,
	74	+ temperature=temperature,
	75	+ )
	76	+ self._last_usage = {
	77	+ "input_tokens": (getattr(response.usage, "prompt_tokens", 0) or 0)
	78	+ if response.usage
	79	+ else 0,
	80	+ "output_tokens": (getattr(response.usage, "completion_tokens", 0) or 0)
	81	+ if response.usage
	82	+ else 0,
	83	+ }
	84	+ return response.choices[0].message.content or ""
	85	+
	86	+ def analyze_image(
	87	+ self,
	88	+ image_bytes: bytes,
	89	+ prompt: str,
	90	+ max_tokens: int = 4096,
	91	+ model: Optional[str] = None,
	92	+ ) -> str:
	93	+ model = model or self._default_vision_model
	94	+ if not model:
	95	+ raise RuntimeError(
	96	+ "No Ollama vision model available. Install a multimodal model: ollama pull llava"
	97	+ )
	98	+ b64 = base64.b64encode(image_bytes).decode()
	99	+ response = self.client.chat.completions.create(
	100	+ model=model,
	101	+ messages=[
	102	+ {
	103	+ "role": "user",
	104	+ "content": [
	105	+ {"type": "text", "text": prompt},
	106	+ {
	107	+ "type": "image_url",
	108	+ "image_url": {"url": f"data:image/jpeg;base64,{b64}"},
	109	+

	--- a/video_processor/providers/ollama_provider.py
	+++ b/video_processor/providers/ollama_provider.py
	@@ -0,0 +1,109 @@

	--- a/video_processor/providers/ollama_provider.py
	+++ b/video_processor/providers/ollama_provider.py
	@@ -0,0 +1,109 @@
1	"""Ollama provider implementation using OpenAI-compatible API."""
2
3	import base64
4	import logging
5	import os
6	from pathlib import Path
7	from typing import Optional
8
9	import requests
10	from openai import OpenAI
11
12	from video_processor.providers.base import BaseProvider, ModelInfo
13
14	logger = logging.getLogger(__name__)
15
16	# Known vision-capable model families (base name before the colon/tag)
17	_VISION_FAMILIES = {
18	"llava",
19	"llava-llama3",
20	"llava-phi3",
21	"llama3.2-vision",
22	"moondream",
23	"bakllava",
24	"minicpm-v",
25	"deepseek-vl",
26	"internvl2",
27	}
28
29	DEFAULT_HOST = "http://localhost:11434"
30
31
32	class OllamaProvider(BaseProvider):
33	"""Ollama local LLM provider via OpenAI-compatible API."""
34
35	provider_name = model = model or self._dprompt_tokens", 0) self.host ULT_HOST)
36	model = model or secmethod
37	def is_available(host: Optional[str] = None) -> bool:
38	"""Check if an Ollama server is running and reachable."""
39	host = host or os.getenv("OLLAMA_HOST", DEFAULT_HOST)
40	try:
41	resp = requests.get(f"{host}/api/tags", timeout=3)
42	return resp.status_code == 200
43	except Exception:
44	return False
45
46	@property
47	def _default_model(self) -> str:
48	models = self._get_models()
49	for m in models:
50	if "chat" in m.capabilities:
51	return m.id
52	return "llama3.2:latest"
53
54	@property
55	def _default_vision_model(self) -> Optional[str]:
56	models = self._get_models()
57	for m in models:
58	if "vision" in m.capabilities:
59	return m.id
60	return None
61
62	def _get_models(self) -> list[ModelInfo]:
63	if self._models_cache is None:
64	self._models_cache = self.list_models()
65	return self._models_cache
66
67	def chat(
68	self,
69	messages: list[ model = model or self._dprompt_tokens", 0) self.host ULT_HOST)
70	model = model or setions.create(
71	model=model,
72	messages=messages,
73	max_tokens=max_tokens,
74	temperature=temperature,
75	)
76	self._last_usage = {
77	"input_tokens": (getattr(response.usage, "prompt_tokens", 0) or 0)
78	if response.usage
79	else 0,
80	"output_tokens": (getattr(response.usage, "completion_tokens", 0) or 0)
81	if response.usage
82	else 0,
83	}
84	return response.choices[0].message.content or ""
85
86	def analyze_image(
87	self,
88	image_bytes: bytes,
89	prompt: str,
90	max_tokens: int = 4096,
91	model: Optional[str] = None,
92	) -> str:
93	model = model or self._default_vision_model
94	if not model:
95	raise RuntimeError(
96	"No Ollama vision model available. Install a multimodal model: ollama pull llava"
97	)
98	b64 = base64.b64encode(image_bytes).decode()
99	response = self.client.chat.completions.create(
100	model=model,
101	messages=[
102	{
103	"role": "user",
104	"content": [
105	{"type": "text", "text": prompt},
106	{
107	"type": "image_url",
108	"image_url": {"url": f"data:image/jpeg;base64,{b64}"},
109

PlanOpticon

Keyboard Shortcuts